We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CV

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computer Vision and Pattern Recognition

Title: Incremental Learning Techniques for Semantic Segmentation

Abstract: Deep learning architectures exhibit a critical drop of performance due to catastrophic forgetting when they are required to incrementally learn new tasks. Contemporary incremental learning frameworks focus on image classification and object detection while in this work we formally introduce the incremental learning problem for semantic segmentation in which a pixel-wise labeling is considered. To tackle this task we propose to distill the knowledge of the previous model to retain the information about previously learned classes, whilst updating the current model to learn the new ones. We propose various approaches working both on the output logits and on intermediate features. In opposition to some recent frameworks, we do not store any image from previously learned classes and only the last model is needed to preserve high accuracy on these classes. The experimental evaluation on the Pascal VOC2012 dataset shows the effectiveness of the proposed approaches.
Comments: 8 pages, 3 figures, 4 tables
Subjects: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Image and Video Processing (eess.IV)
Journal reference: International Conference on Computer Vision (ICCV), Workshop on Transferring and Adapting Source Knowledge in Computer Vision (TASK-CV) 2019
Cite as: arXiv:1907.13372 [cs.CV]
  (or arXiv:1907.13372v4 [cs.CV] for this version)

Submission history

From: Umberto Michieli [view email]
[v1] Wed, 31 Jul 2019 09:00:15 GMT (2149kb,D)
[v2] Tue, 6 Aug 2019 11:59:30 GMT (2149kb,D)
[v3] Thu, 8 Aug 2019 12:43:59 GMT (2149kb,D)
[v4] Tue, 17 Sep 2019 07:33:25 GMT (2148kb,D)

Link back to: arXiv, form interface, contact.